Discrete graphics cards and MXM modules: workload, memory, host fit and lifecycle
A discrete GPU card or module combines a parallel processor, graphics memory and host/display interfaces. Select the exact hardware and software configuration for the workload, then verify power, cooling, mechanical fit and product lifecycle.
Discrete GPUs supplied as PCIe graphics cards and embedded GPU modules (MXM) for workstations, industrial and embedded computers. Graphics integrated into CPUs, complete GPU servers and the GPU die as a bare semiconductor are outside this page.
CNFX editorial research · Sources checked October 18, 2026 · 4 sources
Separate peak compute, memory capacity and memory-interface bandwidth
NVIDIA’s cited RTX A2000 sheet lists 8.0 TFLOPS single-precision performance and identifies the rate as a peak based on GPU Boost Clock. It lists 6 GB or 12 GB GDDR6 with ECC, a 192-bit memory interface and 288 GB/s bandwidth. Those are identified card specifications, not a measured application throughput or a guarantee that all computation is error-free. [1]
Maxsun’s B580 iCraft listing gives 12 GB GDDR6, a 192-bit memory interface, 19 Gbps graphics memory speed and 456 GB/s bandwidth. Keep the memory-interface rate separate from the host PCIe link or achieved data-transfer rate. [2]
Editorial calculation: assuming 19 Gbit/s is the nominal data rate per data pin, all 192 data bits are counted, and decimal units with 8 bits per byte are used, raw bandwidth = 192 ÷ 8 × 19 = 456 GB/s. This ideal interface-rate calculation ignores protocol, refresh and other overhead; it is not an independently measured transfer rate or application result. The inputs and the listed bandwidth come from Maxsun. [2]
02
Read each power label and physical interface literally
NVIDIA labels the A2000 figure as total board power of 70 W and specifies PCIe 4.0 x16, active cooling and a 2.7 in high × 6.6 in long dual-slot form factor. Confirm the actual chassis clearance and cooling arrangement against that card configuration. [1]
Maxsun lists PCIe 4.0 x8, an 8-pin ATX 12 V connector, 315 × 128 × 44 mm dimensions and 190 W under the label GPU Power Consumption. Do not silently relabel that figure as a total-system rating or the same measurement basis as NVIDIA’s total board power. Obtain the card’s power and transient requirements for the selected host. [2]
For an MXM installation, verify the host/carrier, connector, cooling contact and firmware/driver support together. The nominal module outline alone does not establish compatibility with a particular embedded computer. [4]
03
Check the exact module’s lifecycle before planning supply
ADLINK’s current PCIe graphics-card page lists multiple discrete-card products. Its family-level supply/support language is separate from the lifecycle status of an individual old module. [3]
The cited EGX-MXM-P3000 page explicitly marks the product END OF LIFE. It lists a last-buy date of 14 June 2021 and last shipment of 3 January 2022, and names EGX-MXM-RTX3000 as the recommended replacement. The retained historic 5-year availability feature is not evidence of a new supply commitment starting today; the replacement also requires its own host and software qualification. [4]
04
Keep model-specific features and performance conditions together
The P3000’s archived product data identifies an MXM 3.1 Type B form factor of 82 × 105 mm, 6 GB GDDR5 and 3.9 TFLOPS peak FP32. This is historical model information, not a current availability claim or a performance result for another MXM module. [4]
NVIDIA’s sheet separately lists graphics APIs, compute APIs, display outputs and encode/decode resources. Check the current supported driver, operating system and required application version for the actual installation; an API version printed in a dated datasheet does not establish current support for every software stack. [1]
05
Validate the selected card with its intended workload
Use the exact processor/card or module, memory configuration, host interface, cooling and software stack in the acceptance configuration. NVIDIA’s published peak compute and Maxsun’s memory-interface bandwidth describe different limits; neither alone predicts achieved application latency or throughput. [1][2]
Record the workload and precision, usable memory need, ECC configuration where applicable, host fit, power/cooling evidence and lifecycle plan. Request measured results under those conditions rather than converting catalogue peaks into a promised production rate. [3]
06
Send the workload, host and output data together
Paste it into your enquiry below or into an email to a maker, then replace every bracket.
2 companies list this product among what they make. Company figures are quoted from each
company's own website; every card states where the relationship came from.
ADLINK Technology
TBC
Current official site lists PCIe GPU cards and MXM product families. Confirm the specific module lifecycle; the cited P3000 is end of life.
Explore company information and manufacturing capabilities, then discuss the scope of your project with a relevant supplier.
Share this page
Compiled and source-reviewed by the CNFX editorial team.
Each manufacturer above links to the page on that company's own website where the product is listed, with the date we checked it. How we source and review
· Page last updated 13 October 2026
Buyer questions
Does a raw memory-bandwidth calculation measure actual data transfer?
The calculation assumes nominal per-pin data rate, full stated data width and decimal units. It excludes overhead and does not measure PCIe transfer or application throughput. [2]
Does ECC guarantee every workload result is error-free?
The A2000 sheet lists ECC capability. Confirm the supported configuration; that feature alone does not establish correctness of the whole hardware/software computation. [1]
Is the cited P3000 MXM module currently offered with a new five-year lifecycle?
The cited ADLINK page marks it END OF LIFE and provides historic last-buy and shipment dates. Check the replacement’s availability and host qualification separately. [4]
Sources and evidence scope
Sources checked 2026-10-18. NVIDIA’s identified A2000 datasheet, Maxsun’s card specification and ADLINK’s current card lineup and explicitly end-of-life P3000 page. Peak figures, memory-interface rates and lifecycle statements retain their separate meanings.
